qvib.pro
RU

5 open-source tools for AI dev

An LLM gateway for 1600+ models, a self-hosted NotebookLM, pentest agent swarms, a node-based flow builder, and a fast file manager — how to deploy each.

open-source бесплатно MCP профи

Portkey AI Gateway — open-source прослойка с единым API к 1600+ LLM-моделям, автофолбэком и управлением MCP-серверами checked 2026-07-16

Updated: 16.07.2026

Open source ↗

5 open-source tools for AI dev

When you work with several LLM providers or build an agent system, the infrastructure parts — routing requests, handling notes, a visual pipeline builder — usually get solved with paid services. Here are five open-source projects that cover those jobs in open code: deploy once and they keep working without a subscription.

1. Portkey AI Gateway — one API for 1600+ models

github.com/Portkey-AI/gateway is a layer between your code and a wide range of LLM providers behind a single API. The gateway routes requests, falls back automatically when a provider goes down, caches repeated requests and can manage MCP servers.

How to use it: deploy it as a self-hosted service or use the cloud option (Hosted Gateway). In your code the provider call is replaced with a call to the gateway — if the primary provider is unavailable, the request goes to the backup automatically with no changes to your application logic.

2. open-notebook — your own NotebookLM with local models

github.com/lfnovo/open-notebook is a self-hosted alternative to Google NotebookLM supporting 18+ AI providers, local models included. You upload PDFs, video, audio or websites — the system answers questions about the material, generates podcasts and takes notes with AI help.

How to install: the project deploys via Docker Compose, and the provider configuration (cloud API or local model) is set in the settings after the first launch. It is not tied to a single cloud service — you can keep all data and models on your own infrastructure.

3. T3MP3ST — a team of AI agents for pentesting

github.com/elder-plinius/T3MP3ST is an automated pentesting framework with several AI agents: the agents divide the roles themselves — recon, target scanning, vulnerability hunting, documenting results — from the start through to the final report.

How to use it as a reference: even outside a security context, T3MP3ST is a working example of a multi-agent architecture where a task is split into roles and the agents coordinate with each other. The same role-splitting pattern applies to any multi-step agent pipeline.

4. xyflow — a node builder for your own workflow editor

github.com/xyflow/xyflow is a library for building visual editors with nodes and connections, available for React and Svelte. Well-known workflow editors like n8n and LangFlow are built on the same foundation.

How to use it: you add it like any component library — you define the set of node types, the connection logic between them and the rendering. A good fit when you need your own pipeline builder (agent chains, automation workflows) inside your product rather than an off-the-shelf external solution.

5. superfile — a fast file manager for the terminal

github.com/yorukot/superfile gives you a modern TUI for the terminal: tabs, file previews, hotkeys and themes. It is written in Go and is noticeably faster than the classic mc and ranger in everyday file work.

How to install: through a package manager (Homebrew, Go install or a prebuilt binary from the GitHub releases) — no building from source. Once installed it launches with spf right in the terminal.

How these fit into one stack

A typical combination: Portkey as the single entry point to every LLM provider in the project, xyflow as the visual layer for agent pipelines on top of that gateway, open-notebook as the working knowledge base for the project. T3MP3ST and superfile are more standalone tools: one for security, the other for everyday terminal work.

How we do it

In the qvib.pro knowledge base, the cards on LLM gateways and self-hosted alternatives spell out exactly where each tool slots in alongside Claude — for example, how to configure Portkey as a fallback across several providers inside one agent project. The qvib engine is a living example of a product that uses similar routing and agent architecture principles internally.

  • An overview of LLM gateways and routing: qvib.pro/arsenal/obzor-llm-geytveev-i-routing/
  • Self-hosted alternatives to AI services: qvib.pro/arsenal/self-hosted-alternativy-ai-servisam/
  • The "AI agents" section: qvib.pro/arsenal/ai-agenty/

Читать по-русски →